Figure 3: Associative-memory signature as theta pattern similarity.

(a) Snapshots of spatial patterns of theta activity at the time of the highest PSI value. The normalized mean power of theta activity during a 500 ms time window centred at 332 ms after the cue onset is colour-plotted for each channel. Inset: Time course of PSI. Grey shading indicates the 99.99% CI corrected for the number of time points. (b) Schemas of the pair decoding. (top) An example of associated visual stimulus pair (A1 and A2) and their accompanying theta patterns (schematic illustrations). (bottom) A machine-learning decoder is trained on theta patterns evoked by stimulus A1, and then tested on theta patterns evoked by stimulus A2 to correctly predict the label of the test data (see also Methods). (c) Time courses of the pair-decoding accuracy. Peak accuracies were significantly above chance (20%, gray lines; n=2 monkeys, permutation test, P<0.001, corrected for the number of time points). (d) Confusion matrix of the pair decoding. The elements along the left-to-right diagonal axis correspond to correct predictions for stimuli that were actually paired together. Trials from two monkeys were merged. (e) The pair-decoding accuracy in each brain areas within the recording site. Brain areas other than ENT show significantly higher accuracy than the chance-level (n=2 monkeys, permutation test, ***P<0.001, **P<0.01, *P<0.05, corrected for multiple comparison). The short green lines indicate 5% significance levels for the permutation tests. The grey line indicates the chance-level (20%). ALL: all channels, CI, confidence interval.